Gastrointestinal Imaging • Original Research

Hara et al.
CT Iterative Reconstruction Technique

Gastrointestinal Imaging
Original Research

Iterative Reconstruction Technique

for Reducing Body Radiation Dose
at CT: Feasibility Study
Amy K. Hara1 OBJECTIVE. The purpose of this study was to evaluate the image noise, low-contrast
Robert G. Paden resolution, image quality, and spatial resolution of adaptive statistical iterative reconstruction
Alvin C. Silva in low-dose body CT.
Jennifer L. Kujak MATERIALS AND METHODS. Adaptive statistical iterative reconstruction was used
Holly J. Lawder to scan the American College of Radiology phantom at the American College of Radiology
reference value and at one-half that value (12.5 mGy). Test objects in low- and high-contrast
William Pavlicek
and uniformity modules were evaluated. Low-dose CT with adaptive statistical iterative re-
Hara AK, Paden RG, Silva AC, Kujak JL, construction was then tested on 12 patients (seven men, five women; average age, 67.5 years)
Lawder HJ, Pavlicek W who had previously undergone routine-dose CT. Two radiologists blinded to scanning tech-
nique evaluated images of the same patients obtained with routine-dose CT and low-dose CT
with and without adaptive statistical iterative reconstruction. Image noise, low-contrast reso-
lution, image quality, and spatial resolution were graded on a scale of 1 (best) to 4 (worst).
Quantitative noise measurements were made on clinical images.
RESULTS. In the phantom, low- and high-contrast and uniformity assessments showed
no significant difference between routine-dose imaging and low-dose CT with adaptive sta-
tistical iterative reconstruction. In patients, low-dose CT with adaptive statistical iterative re-
construction was associated with CT dose index reductions of 32–65% compared with routine
imaging and had the least noise both quantitatively and qualitatively (p < 0.05). Low-dose CT
with adaptive statistical iterative reconstruction and routine-dose CT had identical results for
low-contrast resolution and nearly identical results for overall image quality (grade 2.1–2.2).
Spatial resolution was better with routine-dose CT (p = 0.004).
CONCLUSION. These preliminary results support body CT dose index reductions of
32–65% when adaptive statistical iterative reconstruction is used. Studies with larger statisti-
cal samples are needed to confirm these findings.

Keywords: body CT, low-dose CT, radiation dose
DOI:10.2214/AJR.09.2397
Received January 15, 2009; accepted after revision
February 24, 2009.
1 than two thirds of the total radiation dose as-
All authors: Department of Diagnostic Radiology, Mayo
Clinic Arizona, 13400 E Shea Blvd., Scottsdale, AZ 85259.
Address correspondence to A. K. Hara.
AJR 2009; 193:764–771
0361–803X/09/1933–764
© American Roentgen Ray Society
T
he explosive growth of CT can
be attributed to its wide avail-
ability, speed, and diagnostic
benefits. In a 2007 report [1] it
was estimated that more than 68.7 million
CT examinations are performed each year in
the United States, a dramatic upsurge com-
pared with the 3 million performed in 1980.
Although it accounts for only 11–13% of ra-
diologic examinations performed overall in
the United States, CT is responsible for more
sociated with medical imaging [2, 3]. Public
concern with radiation exposure escalated
when a widely publicized article [4] claimed
that the estimated cancer risk in the United
States attributable to CT radiation has grown
from 0.4% to 1.5–2.0% owing to the substan-
tial increase in use of CT.
Dose reduction with CT has been limited
because the current CT reconstruction algo-
rithm (filtered back projection [FBP]) does
not produce consistently diagnostic images if
tube current is substantially reduced. Iterative
reconstruction is a reconstruction algorithm
whereby image data are corrected with an as-
sortment of models. Although new to CT, it-
erative reconstruction is widely used in PET
and was used when CT was introduced [5]. A
current limitation of iterative reconstruction,
however, is the long computing time. There-
fore, a modified and computationally faster
iterative reconstruction technique, adaptive
statistical iterative reconstruction, was devel-
oped in which only one corrective model is
used to address image noise. This technique
is used to solve one of the primary problems
of dose reduction for CT with FBP: increased

764 AJR:193, September 2009

 CT Iterative Reconstruction Technique

noise with decreased radiation dose. The pur-
pose of this study was to determine the fea-
sibility of adaptive statistical iterative recon-
struction for low-dose body CT through an
evaluation of image noise, low-contrast res-
olution, image quality, and spatial resolution
both in a phantom and in patients.
Materials and Methods
Study Design
All examinations were performed on a 64-MDCT
scanner (CT750 HD, GE Healthcare). This retro-
spective HIPAA-compliant study was approved by
the institutional review board of our institution.
Adaptive Statistical Iterative Reconstruction
In computation with iterative reconstruction,
the image has an initial condition of values, which
are iteratively optimized according to the rules of
the model. The FBP image is used for the initial
condition in adaptive statistical iterative recon-
struction (the initial value of each pixel) for the
following reasons: it is presumably close to the fi-
nal optimized solution (lessening the need for it-
erations); it is a valid indicator of specific-slice
image noise; and it can be quickly obtained. For
modeling and use of iterative reconstruction, min-
imum convergence is achievable with the adaptive
statistical iterative reconstruction model (Fig. 1).
A fully converged 100% adaptive statistical itera-
tive reconstruction image, however, tends to have
a noise-free appearance with unusually homoge-
neous attenuation. Because some noise is inherent
in CT, use of 100% adaptive statistical iterative re-
construction may not be immediately appealing to
most radiologists. However, a linear mixture of the
original FBP and the full adaptive statistical itera-
tive reconstruction images can result in a blended
image with markedly decreased noise that retains
a more typical CT appearance. This blended im-
age can be adjusted from 1% to 100% in adaptive
statistical iterative reconstruction. A mathematic
description of adaptive statistical iterative recon-
struction is shown in Appendix 1.
Phantom Study
The American College of Radiology (ACR) CT
phantom (Gammex 464, Gammex) was scanned
twice, once with the ACR reference values (www.
acr.org/accreditation/computed/ct_reqs.aspx) and
then at one-half this value (12.5 mGy). Helical
scanning was performed in the manner required
for submission of images for scanner accreditation.
Low-contrast resolution, high-contrast resolution,
and uniformity modules were imaged, and these
test objects were evaluated by CT physicists not
blinded to scanning technique. Radiation dose and
noise estimates were made in accordance with ACR
protocol. Images were reconstructed with both FBP
and multiple values of adaptive statistical iterative
reconstruction ranging from 10% to 100%.
Patient Study
The study cohort consisted of 12 patients (seven
men, five women; average age, 67.5 years; range,
53–86 years) who consecutively underwent low-
dose CT and who had undergone routine-dose CT
of the same region (abdomen or abdomen and pel-
vis) within an average of 10.1 months (range, 3
days–5 years) before low-dose CT. The compari-
sons were matched for IV contrast enhancement
and imaging phase. In six comparisons, only the
abdomen had been imaged, and in six, only the ab-
domen and pelvis. Ten of the 12 comparisons had
been performed with IV contrast enhancement
(nine venous phase, one arterial phase). In the
other two examinations, CT was unenhanced. For
routine-dose CT, the peak kilovoltage had been
140 kVp in eight examinations and 120 kVp in four.
Seven of the CT examinations had been performed
with dose modulation software with variable tube
current at a slice thickness of 5 mm in two exami-
nations, 3.75 mm in five, and 3 mm in five exam-
inations. In most cases, 64-MDCT scanners had
been used (three, VCT, GE Healthcare; six, Sen-
sation 64, Siemens Healthcare). The other three
examinations were performed with a 16-MDCT
scanner (Sensation 16, Siemens Healthcare).
Technique of Low-Dose CT With Adaptive
Statistical Iterative Reconstruction
Low-dose CT was performed with the follow-
ing parameters: fixed noise index, 30.9; collima-
tion, 0.625 mm; reconstruction slice thickness,
3.75 mm; tube potential, 120 kV; variable tube cur-
rent determined by x, y, z-axis dose modulation;
gantry rotation time, 0.5 second. The CT dose was
reduced through an increase in accepted noise in-
dex for the study from 22.1 to 30.9. With the dose
modulation software of the scanner, the tube cur-
rent was automatically reduced to match the ac-
ceptable noise index. The dose varied with patient
size; that is, larger patients needed a higher tube
current for maintenance of the desired noise index
than did thinner patients. All 12 low-dose CT ex-
aminations were reconstructed twice, once with
FBP and once with 40% adaptive statistical itera-
tive reconstruction. The 40% level was chosen on

A B C
Fig. 1—Production of adaptive statistical iterative reconstruction image.
A, Filtered back projection image obtained at 120 kVp and 300 mA at 12.5 mGy (half dose).
B, Image from 100% adaptive statistical iterative reconstruction generated through multiple iterations in accordance with rules of noise reduction model.
C, Linear combination of A and B produces blended image (50% adaptive statistical iterative reconstruction), which has less noise than filtered back projection image but
without artifactual smoothing of 100% adaptive statistical iterative reconstruction image.

AJR:193, September 2009 765

 Hara et al.

30.00
10% increments, the result was a linear de-
Noise (standard deviation)
Full Dose
Half Dose
crease in noise as measured with SD (Fig. 2).
25.00 For full dose scanning, at 0% adaptive statis-
20.00 tical iterative reconstruction, the SD (noise)
was 20. At 100% adaptive statistical iterative
15.00
reconstruction, the noise was minimum (SD
10.00 4), an approximately 75% reduction of noise
5.00 from the original data. At approximately
50% adaptive statistical iterative reconstruc-
0.00
0 10 20 30 40 50 60 70 80 90 100 tion, the noise was approximately one-half
% Adaptive Statistical Iterative Reconstruction
that of a full-dose FBP image.
When the phantom was scanned at 50%
Fig. 2—Noise reduction in images reconstructed with adaptive statistical iterative reconstruction in phantom. lower dose, the noise as measured with SD
Graph shows linear decrease in image noise (SD) as percentage adaptive statistical iterative reconstruction was 1.4 times greater (28.6 vs 20.4) with 0%
increases. Images acquired with 50% dose reduction (half dose) have 1.4 times SD value (28.57 compared with adaptive statistical iterative reconstruction. At
20.39) without adaptive statistical iterative reconstruction. Reconstructing images with 30% adaptive statistical
iterative reconstruction for half-dose acquisitions produces images with noise nearly equivalent to that of full- 30% adaptive statistical iterative reconstruc-
dose images without adaptive statistical iterative reconstruction (double arrow) (SD 20.52 compared with 20.39). tion, the noise was equivalent to that of a full-
dose FBP image, and further reductions in
the basis of results of the phantom analysis, which Image noise, image quality, low-contrast resolu- noise occurred as percentage adaptive statisti-
indicated that 40% adaptive statistical iterative re- tion, and spatial resolution were graded on a scale cal iterative reconstruction was increased.
construction should produce a diagnostically ac- from 1 (best) to 4 (worst). A score of 1 meant that Comparison of low-contrast images
ceptable image with less noise than a full-dose the image was better than expected at routine-dose showed comparable appearance of the ACR-
FBP image. CT, 2 meant the image was equivalent to that ex- required 6-mm objects at both routine-dose
pected at routine-dose CT, 3 meant the image was CT with FBP and low-dose CT with adaptive
Dose Comparison worse than expected at routine-dose CT, and 4 statistical iterative reconstruction (Fig. 3).
Volume CT dose index (CTDI) and dose–length meant the image was nondiagnostic. The readers Comparison of the high-contrast object
product (DLP) were compared for low-dose (n = independently assessed image noise, image qual- (12 line pairs/cm) showed that adaptive sta-
12) and routine-dose (n = 12) CT examinations. ity, and low-contrast resolution. Readers were in- tistical iterative reconstruction was compa-
For comparison of radiation doses, the patients structed to assess low-contrast resolution by eval- rable with FBP in terms of spatial resolution
were divided into three groups based on body uating the conspicuity of the hepatic veins within and easily exceeded the spatial resolution re-
mass index (BMI) (weight in kilograms divided by the liver or solid organ cysts. Spatial resolution was quirement of 6 line pairs/cm required for ac-
height squared in meters): greater than 25 (n = 3), assessed through consensus evaluation by grading creditation (Fig. 4).
20–24.9 (n = 6), and less than 20 (n = 3). of the sharpness of the hepatic or renal edges. Uniformity was maintained at low-dose CT
with adaptive statistical iterative reconstruc-
Quantitative Analysis Results tion (the maximum deviation between the cen-
Two abdominal imaging fellows not involved Phantom Study tral ROI and peripheral ROIs was less than
in qualitative data analysis made quantitative When adaptive statistical iterative recon- 5 HU) and was within ACR specifications
noise measurements on a total of 36 data sets: 12 struction was applied to the FBP image in [6]. Uniformity measurements on routine-
low-dose CT without adaptive statistical iterative
reconstruction, 12 low-dose CT with adaptive sta-
tistical iterative reconstruction, and 12 routine-
dose comparison CT. Noise measurements were
made by recording the SD in an identically sized
2,500-mm 2 region of interest (ROI) placed 5 mm
outside the anterior abdominal wall at the level of
the umbilicus.

Qualitative Analysis
Qualitative image analysis was performed by
two board-certified and fellowship-trained ab-
dominal radiologists with 8 and 10 years of CT
experience. The 36 data sets were randomized and
deidentified so the readers were unaware of the
postprocessing algorithm and dose. Only the axial
images were displayed on a PACS (Centricity ver- A B
sion 2.1, GE Healthcare). All data sets were dis-
Fig. 3—Low-contrast objects of comparable quality.
played at soft-tissue settings (window, ~ 400 HU; A, Filtered back projection image obtained with 25-mGy routine body image protocol.
level, ~ 40). B, Adaptive statistical iterative reconstruction image obtained at 50% reduced dose (12.5 mGy).

766 AJR:193, September 2009

 CT Iterative Reconstruction Technique

dose CT with adaptive statistical iterative re-

construction had the least visually assessed
image noise, and the average of the two read-
ers was significantly better compared with
those for low-dose CT without adaptive sta-
tistical iterative reconstruction (p < 0.0001)
and routine-dose CT (p = 0.01) (Table 3). Av-
eraged scores from both readers in the com-
parison of routine-dose CT and low-dose CT
with adaptive statistical iterative reconstruc-
tion were equivalent or nearly equivalent for
image quality and low-contrast resolution.
Low-dose CT with adaptive statistical iter-
ative reconstruction was significantly better
A B than low-dose CT without it in low-contrast
resolution (p = 0.01) and image quality (p =
Fig. 4—High-contrast objects. 0.0002). The consensus score for spatial res-
A, Filtered back projection image obtained at routine dose of 25 mGy shows 12 line pairs/cm.
B, CT scan obtained with adaptive statistical iterative reconstruction at low dose of 12.5 mGy shows 12 line olution was significantly better for routine-
pairs/cm despite more image degradation than in A. dose CT compared with low-dose CT with-
out (p = 0.002) and with (p = 0.004) adaptive
dose FBP images and low-dose CT images CTDI and DLP of 65% compared with only statistical iterative reconstruction.
with adaptive statistical iterative reconstruc- 29–35% for patients with a BMI was greater Three comparisons had identical scanning
tion were nearly identical (Fig. 5). than 25. parameters (peak kilovoltage, slice thickness,
Quantitative comparisons—Quantitative scanner model), and the results were nearly
Patient Study ROI noise measurements were highest for identical to those in the overall group anal-
Dose comparison—Overall, use of a low low-dose CT without adaptive statistical it- ysis. The only difference was that the aver-
dose reduced the CTDI 47% and the DLP erative reconstruction (average, 35; range, aged image quality grade was slightly higher
44% (Table 1). For low-dose CT, the average 11–52). The noise indexes were nearly iden- for low-dose CT with adaptive statistical it-
CTDI was 12 compared with 22 for routine- tical for low-dose CT with adaptive statis- erative reconstruction (2.0) than for routine-
dose CT. The average DLP was 470 mGy·cm tical iterative reconstruction (average, 31; dose CT (2.2). In the overall analysis, rou-
for low-dose CT compared with 894 mGy·cm range, 7–51) and routine-dose CT (average, tine-dose CT (2.1) had slightly better overall
for routine-dose CT. Not surprisingly, when 32; range, 10–45). image quality (low-dose CT with adaptive
the 12 examinations were compared on the Qualitative comparisons—Low-dose CT statistical iterative reconstruction, 2.2).
basis of BMI (Table 2), the percentage re- without adaptive statistical iterative recon-
ductions in CTDI and DLP increased as BMI struction had the worst scores for visually Discussion
decreased. Therefore, patients with a BMI assessed image noise, image quality, spatial Several approaches have been used in the
less than 20 had percentage reductions in resolution, and low-contrast resolution. Low- effort to minimize radiation dose at CT [7].
These include use of automated tube current
modulation [8–12], noise reduction filters [13,
14], and low-dose protocols for specific clini-
cal indications. With automated tube modula-
tion techniques, CTDI reductions of 40–60%
have been achieved without compromise of
image quality and are now routinely used on
most scanners [7, 12, 15]. Further dose re-
ductions have been achieved mainly by im-
plementation of specific low-dose protocols
for indications such as renal stones and co-
lonic polyps [16–19]. These protocols do not
include IV contrast administration, and us-
ing them can reduce dose more than 50% if
image quality outside the area of interest is
sacrificed. The renal stone and CT colonog-
A B
raphy protocols are not widely implemented
Fig. 5—Uniformity module in phantom. for routine body CT, however, because most
A and B, Filtered back projection image obtained at routine dose of 25 mGy (A) has uniformity nearly identical routine body CT examinations are performed
to that of low-dose (12.5 mGy) CT scan obtained with adaptive statistical iterative reconstruction (B). Central
region of interest (ROI) has attenuation between +3 and –3 HU. Maximum deviation between central ROI and with IV contrast material. In addition, radi-
peripheral ROIs must be less than 5 HU. ologists perform many of these examinations

AJR:193, September 2009 767

 Hara et al.

TABLE 1:  Comparison of Low-Dose CT and Routine-Dose CT of Same Patients

Dose–Length Product (mGy⋅cm) CT Dose Index
Body
Patient IV Contrast Mass Percentage Percentage
No. Type of Examination Enhancement Index Routine Dose Low Dose Reduction Routine Dose Low Dose Reduction
1 Abdomen Yes 34 707 441 38 27 17 37
2 Abdomen No 30 1,008 773 23 31 20 37
3 Abdomen, pelvis Yes 28 1,209 886 27 26 18 32
4 Abdomen Yes 25 376 305 19 14 9 35
5 Abdomen Yes 25 848 502 41 26 15 44
6 Abdomen, pelvis No 22 921 549 40 20 11 43
7 Abdomen, pelvis Yes 22 860 548 36 18 11 39
8 Abdomen Yes 20 396 197 50 13 6 52
9 Abdomen, pelvis Yes 20 1,128 451 60 21 8 62
10 Abdomen Yes 19 353 114 68 14 5 65
11 Abdomen, pelvis Yes 19 1,198 442 63 26 9 65
12 Abdomen, pelvis Yes 18 1,073 430 60 22 8 62
Average 24 840 470 44 22 11 48

without a clear idea of the diagnosis and fear
missing an important finding owing to poor
image quality. A method of reducing dose at
routine abdominal CT has been lacking.
In this preliminary study, we evaluated an
alternative approach to reducing CT radiation
dose for routine abdominal CT that has not
been previously available for clinical use, to
our knowledge. This approach, adaptive sta-
tistical iterative reconstruction, is a unique
CT reconstruction algorithm compared with
the only one previously available (FBP). Un-
like with FBP, with adaptive statistical iter-
ative reconstruction, it is not assumed that
noise is evenly distributed across the entire
image. Instead, matrix algebra is used to se-
lectively identify and then subtract noise from
the image with a mathematic model. The re-
sult is a less noisy image. The ability to selec-
tively reduce image noise allows generation
of a higher-quality image at a lower radiation
dose with adaptive statistical iterative recon-
struction than with FBP techniques.
For this study, we sought to determine the
feasibility of using adaptive statistical itera-
TABLE 2:  Subset Comparison According to Body Mass Index
Average Dose–Length Product
(mGy⋅cm)
Body
Mass No. of Routine
Index Patients Dose Low Dose
> 25 3 975 700
20–24.9 6 755 425
< 20 3 875 328
tive reconstruction for low-dose body CT of
both a phantom and patients. An ACR phan-
tom was used to validate the use of adaptive
statistical iterative reconstruction. We found
that use of a dose reduced by 50% (12.5
mGy) and adaptive statistical iterative recon-
struction yielded low- and high-contrast res-
olution and image uniformity within ACR
specifications. Images obtained with the
technique easily exceeded the spatial reso-
lution requirement for ACR accreditation.
We are applying for ACR accreditation using
this technique.
After the validation, low-dose CT with
adaptive statistical iterative reconstruction
was tested in a group of 12 patients who
had undergone routine-dose imaging of the
abdomen or abdomen and pelvis. This pre-
liminary evaluation showed use of adaptive
statistical iterative reconstruction reduces
readers’ perception of image noise in spite
of a decrease in CTDI that was as high as
65% (Fig. 6). Actual measurements of image
noise were nearly identical for low-dose CT
with adaptive statistical iterative reconstruc-
Average Volume CT Dose Index
Percentage Routine Percentage
Reduction Dose Low Dose Reduction
29 28 18 35
41 19 10 46
64 21 7 64
tion and routine-dose CT. Not surprisingly,
when CTDI was reduced without application
of adaptive statistical iterative reconstruc-
tion, both readers found the images noisier
than images from full-dose routine examina-
tions, and quantitative noise measurements
also were higher. These results support the
ability of adaptive statistical iterative re-
construction to allow substantial reduc-
tions in radiation dose without the compro-
mise in image quality due to noise that once
was so troublesome. Furthermore, reader as-
sessments of overall image quality and low-
contrast resolution were nearly identical for
low-dose CT scans with adaptive statistical
iterative reconstruction and routine-dose im-
ages, further supporting the potential of this
technique for routine imaging.
Spatial resolution was the only parameter
that was worse with adaptive statistical iter-
ative reconstruction than with routine-dose
imaging in this pilot study (Fig. 7), although
the actual score (2.5) meant the image had
only slightly lower spatial resolution than was
expected for routine-dose CT. In the review
of adaptive statistical iterative reconstruc-
tion images with low scores, the main differ-
ences were slightly decreased sharpness of
cyst edges and a mildly irregular or jagged
margin of solid organs. These imperfections
did not appear to affect the diagnostic value
of the image itself. This factor was evaluat-
ed only on axial images. Coronal and sagit-
tal multiplanar reformations were not avail-
able or included in the evaluation. This issue
will be evaluated in a future study involving

768 AJR:193, September 2009

 CT Iterative Reconstruction Technique

TABLE 3:  Visual Assessments by Two Readers

Image Noise Low-Contrast Resolution Overall Image Quality Spatial Resolution
Low Dose Low Dose Low Dose Low Dose
Non- Routine Non- Routine Non- Routine Non- Routine
Reader ASIR ASIR Dose ASIR ASIR Dose ASIR ASIR Dose ASIR ASIR Dose
A 2.8 1.6 2.2 2.3 2.2 2.2 2.5 2.3 2.3 NA
B 2.8 1.6 2.3 2.5 2.0 2.1 3.0 2.2 1.9 NA
Averagea 2.8 1.6 2.2 2.4 2.1 2.1 2.8 2.2 2.1 2.6 2.5 1.9
Note—Values are qualitative grading scale: 1, better than routine-dose CT; 2, similar to routine-dose CT; 3, worse than routine-dose CT; 4, nondiagnostic. ASIR = low-dose
CT with adaptive statistical iterative reconstruction, NA = not applicable.
aLow-dose CT with adaptive statistical iterative reconstruction was significantly better than routine-dose CT for image noise (p = 0.01). Low-dose CT with ASIR was

significantly better than low-dose CT without ASIR for image noise, low-contrast resolution, and overall image quality (p < 0.01). Routine-dose CT was significantly better
than low-dose CT with or without ASIR for spatial resolution (p ≤ 0.004).

A B C
Fig. 6—57-year-old woman with body mass index of 18.
A–C, Low-dose CT scan obtained at 120 kVp, 3.75-mm slice thickness, and CT dose index (CTDI) of 8 without adaptive statistical iterative reconstruction (A) has more
image noise in liver than low-dose CT scan with adaptive statistical iterative reconstruction (B) and routine-dose CT scan (140 kVp; 3-mm slice thickness; CTDI, 22) (C). B
and C have nearly identical image quality.

a group of patients with lesions to determine
whether diagnostic confidence is affected.
Future releases of adaptive statistical itera-
tive reconstruction software also may help to
resolve this issue.
The degree of dose reduction was great-
est for patients with a lower BMI. In patients
with a BMI less than 20, the average CTDI
dose reduction was 64% compared with 35%
for patients with a BMI greater than 25. That
adaptive statistical iterative reconstruction
allows dose reductions for smaller patients
may help with pediatric imaging, which was
not evaluated in this study. The idea of dose
reduction in CT of even larger patients by use
of adaptive statistical iterative reconstruction
is encouraging because the number of obese
patients in the United States continues to in-
crease [20].
These preliminary results may help to en-
courage more widespread use of low-dose CT
protocols. In our practice, we have instituted
use of low-dose CT with adaptive statistical
iterative reconstruction for all body CT per-
formed on scanners with this reconstruction
algorithm available. For our scanners that do
not have adaptive statistical iterative recon-
struction capability, we have used these results
to investigate ways to reduce our standard-
dose CT protocols, particularly in imaging of
smaller patients. Studies in neurologic, mus-
culoskeletal, chest, and cardiac CT are ongo-
ing to determine whether low-dose protocols
can be used in these areas.
Even more aggressive reductions in radi-
ation dose may be possible in the future. In
effect, scanning may be performed at doses
low enough to render images nearly nondiag-
nostic but with advanced iterative reconstruc-
tion techniques to return image quality to an
acceptable level. Currently, the use of itera-
tive reconstruction at CT is limited by long
reconstruction times. As hardware and soft-
ware improve, more complex iterative recon-
struction algorithms may be used clinically,
resulting in even greater improvements in im-
age quality. Iterative reconstruction also may
allow routine image reconstruction at thin-
ner slices. Currently, increased noise limits
the evaluation of thin reconstructed images
(< 2.5 mm) in abdominal imaging. With it-
erative reconstruction and adaptive statistical
iterative reconstruction, whether or not radia-
tion dose is reduced, thinner reconstructions
may become diagnostic, improving detection
and characterization of lesions.
This initial evaluation had limitations.
First, because of the retrospective nature of
the study, the low-dose and routine-dose CT
examinations did not have identical scan-
ning parameters. Changes in peak kilovolt-
age and slice thickness can affect noise and
image quality. Results of prospective studies
with similar imaging parameters will be help-
ful for confirming the initial results. Second,
the small sample size had limited power, and
prospective studies with larger samples are
needed. In addition, for the purposes of this
study, we chose an adaptive statistical itera-
tive reconstruction level of 40% because it ap-
proximated the levels in the phantom study. It
is possible that higher levels of adaptive statis-
tical iterative reconstruction may improve re-
sults. Finally, we did not assess lesions spe-
cifically. It is possible that adaptive statistical
iterative reconstruction may affect lesion con-
spicuity and detection, and this factor has to
be assessed with future studies.
We conclude that low-dose body CT with
adaptive statistical iterative reconstruction
has quantitative and qualitative image noise

AJR:193, September 2009 769


A B
Fig. 7—75-year-old man with body mass index of 22.
A–C, Low-dose CT scans without (A) and with (B) adaptive statistical iterative reconstruction (120 kVp; 3.75-mm slice thickness; CT dose index [CTDI], 11) and routine-
dose CT scan (C) (140 kVp; 3-mm slice thickness; CTDI, 20) all show hepatic cysts, but sharpness of cyst edges is best in C. B has least image noise.
and image quality similar to or better than
those of routine-dose CT. Compared with
those of conventional imaging, with adaptive
statistical iterative reconstruction, CTDI and
DLP both were reduced an average of nearly
50% and up to 65% in some patients using
adaptive statistical iterative reconstruction
when compared with routine dose imaging.
The low-dose technique with adaptive sta-
tistical iterative reconstruction met ACR ac-
creditation standards in a phantom analysis.
On the basis of our results, we have imple-
mented this low-dose technique for routine
body CT in our practice. Future studies are
needed to confirm these preliminary results
and determine the effect of adaptive statisti-
cal iterative reconstruction on lesion detec-
tion and conspicuity.
Acknowledgment
We thank Qing Wu for statistical help.
References
1. 2007 CT market summary report. Des Plaines,
IL: IMV Medical Information Division, 2007
2. Kalra MK, Maher MM, Toth TL, et al. Strategies
for CT radiation dose optimization. Radiology
2004; 230:619–628
3. Linton OW, Mettler FA Jr. National conference on
dose reduction in CT, with an emphasis on pediat-
ric patients. AJR 2003; 181:321–329
4. Brenner DJ, Hall EJ. Computed tomography: an
increasing source of radiation exposure. N Engl J
Med 2007; 357:2277–2284
5. Hounsfield GN. Computerized transverse axial
scanning (tomography). Part 1. Description of
770 AJR:193, September 2009
Hara et al.
system. Br J Radiol 1973; 46:1016–1022
6. McCollough CH, Bruesewitz MR, McNitt-Gray
MF, et al. The phantom portion of the American
College of Radiology (ACR) computed tomogra-
phy (CT) accreditation program: practical tips,
artifact examples, and pitfalls to avoid. Med Phys
2004; 31:2423–2442
7. Valentin J; International Commission on Radia-
tion Protection. Managing patient dose in multi-
detector computed tomography (MDCT): ICRP
Publication 102. Ann ICRP 2007; 37:1–79
8. Mulkens TH, Bellinck P, Baeyaert M, et al. Use of
an automatic exposure control mechanism for
dose optimization in multi-detector row CT ex-
aminations: clinical evaluation. Radiology 2005;
237:213–223
9. Smith AB, Dillon WP, Lau BC, et al. Radiation
dose reduction strategy for CT protocols: success-
ful implementation in neuroradiology section.
Radiology 2008; 247:499–506
10. Graser A, Wintersperger BJ, Suess C, Reiser MF,
Becker CR. Dose reduction and image quality in
MDCT colonography using tube current modula-
tion. AJR 2006; 187:695–701
11. Lee CH, Goo JM, Ye HJ, et al. Radiation dose
modulation techniques in the multidetector CT
era: from basics to practice. RadioGraphics 2008;
28:1451–1459
12. McCollough CH, Bruesewitz MR, Kofler JM Jr.
CT dose reduction and dose management tools:
overview of available options. RadioGraphics
2006; 26:503–512
13. Kalra MK, Maher MM, Blake MA, et al. Detection
and characterization of lesions on low-radiation-dose
abdominal CT images postprocessed with noise
reduction filters. Radiology 2004; 232: 791–797
C
14. Kalra MK, Maher MM, Sahani DV, et al. Low-
dose CT of the abdomen: evaluation of image im-
provement with use of noise reduction filters pilot
study. Radiology 2003; 228:251–256
15. Kalra MK, Maher MM, Toth TL, Kamath RS,
Halpern EF, Saini S. Radiation from “extra” im-
ages acquired with abdominal and/or pelvic CT:
effect of automatic tube current modulation. Radi-
ology 2004; 232:409–414
16. Cohnen M, Vogt C, Beck A, et al. Feasibility of
MDCT colonography in ultra-low-dose technique
in the detection of colorectal lesions: comparison
with high-resolution video colonoscopy. AJR
2004; 183:1355–1359
17. Niemann T, Kollmann T, Bongartz G. Diagnostic
performance of low-dose CT for the detection of
urolithiasis: a meta-analysis. AJR 2008; 191:396–
401
18. Poletti PA, Platon A, Rutschmann OT, Schmidlin
FR, Iselin CE, Becker CD. Low-dose versus stan-
dard-dose CT protocol in patients with clinically
suspected renal colic. AJR 2007; 188:927–933
19. Luz O, Buchgeister M, Klabunde M, et al. Evalu-
ation of dose exposure in 64-slice CT colonogra-
phy. Eur Radiol 2007; 17:2616–2621
20. Ford ES, Mokdad AH. Epidemiology of obesity in
the western hemisphere. J Clin Endocrinol Metab
2008; 93:S1–S8
21. Thibault JB, Sauer KD, Bouman CA, Hsieh J. A
three-dimensional statistical approach to im-
proved image quality for multislice helical CT.
Med Phys 2007; 34:4526–4544
22. Zhang Y, Fessier J, Hsieh J. Fast variance image
predictions for quadratically regularized statisti-
cal image reconstruction in fan-beam tomogra-
phy. IEEE Nuclear Science Symp Conf 2005; 4:4
(Appendix appears on next page)
 CT Iterative Reconstruction Technique

APPENDIX 1: Mathematical Description of Adaptive Statistical Iterative Reconstruction

Mathematically, the adaptive statistical iterative reconstruction model [21] is based on matrix algebra whereby the measured value of each
pixel (y) is transformed to a new estimate of the pixel (y'). This pixel value is compared with the ideal answer predicted with the noise model
(A). If needed, another iteration ensues.
xˆ = argmin{L(Ax,
x
y) + αG(x)}

A final pixel value for the adaptive statistical iterative reconstruction image (x̂) results when repeated y' values ultimately converge. G is the
regularization term that enforces smoothing and edge preservation of the data [22].

F O R YO U R I N F O R M AT I O N
PQI Connect is the latest addition to the ARRS Website and serves as a source for information on meeting the
growing demand for quality review programs in today’s radiology practices and facilities. The interactive and
easy-to-navigate site focuses on five critical topics that guide you through news items, relevant articles, and links
to important information on each topic.

AJR:193, September 2009 771